Claude Skills vs Custom GPTs vs Gems: April 2026

Every frontier lab now has its own “app platform” for AI. Anthropic ships Claude Skills. OpenAI ships Custom GPTs. Google ships Gemini Gems. They sound similar but solve very different problems — and picking the wrong one to build on in April 2026 can waste months. Here’s the honest comparison.

Last verified: April 20, 2026

TL;DR

Factor	Winner
Power (what they can do)	Claude Skills
Distribution reach	Custom GPTs
Ease of creation	Gemini Gems
Monetization	Custom GPTs (largest), Skills (best terms)
Code execution	Claude Skills
Enterprise controls	Tie: Skills + Custom GPTs
File knowledge	Tie
Tool/API integration	Custom GPTs (Actions)
MCP native	Claude Skills

What each actually is

Claude Skills

Executable packages that extend Claude’s capabilities. A Skill is a directory with:

• SKILL.md — instructions + description (Claude reads it on-demand)
• Optional references/ — docs Claude can read
• Optional scripts/ — code Claude can execute
• Optional subdirectories with examples, templates, etc.

When a user’s request matches a Skill’s description, Claude loads it. Skills work in Claude Code (CLI) and Claude.ai. They can execute shell commands, run scripts, and use MCP servers.

Launched: Mid-2025, matured through early 2026.

Custom GPTs

User-configurable ChatGPT personas. A Custom GPT has:

• Instructions (system prompt)
• Knowledge files (up to 20)
• Actions (OpenAPI-defined API calls)
• Capabilities toggles (web browsing, image gen, Code Interpreter)

Custom GPTs live in ChatGPT and are discoverable in the GPT Store (launched January 2024, now has millions of GPTs).

Launched: November 2023, monetization in 2024.

Gemini Gems

Personalized Gemini assistants. A Gem has:

• Instructions (persona + task)
• Optional file knowledge
• No tools, no actions, no code execution

Gems are created in Gemini (free or Workspace) and are strictly user-level or Workspace-shared. No marketplace.

Launched: Late 2024, matured through 2025.

Feature comparison

Feature	Claude Skills	Custom GPTs	Gemini Gems
Instructions	✅	✅	✅
File knowledge	✅ (unlimited)	✅ (20 files)	✅ (limited)
Web browsing	✅ (if granted)	✅	✅
Code execution	✅ (shell + scripts)	✅ (Code Interpreter)	❌
External API calls	✅ (MCP)	✅ (Actions)	❌
MCP support	✅ Native	⚠️ Via Actions wrapper	❌
Git-based authoring	✅ (directories)	❌ (UI-only)	❌ (UI-only)
Marketplace	✅	✅ (largest)	❌
Monetization	✅	✅	❌
Enterprise controls	✅	✅	⚠️ (Workspace only)

Code / authoring side-by-side

Claude Skill — file-based

my-skill/
├── SKILL.md
├── references/
│   └── api-docs.md
└── scripts/
    └── lint-and-fix.sh

SKILL.md:

---
name: astro-seo-fix
description: Fix SEO issues in Astro answer pages. Use when user asks to audit or fix SEO on andrew.ooo answer pages.
---

# Astro SEO Fix

Read references/api-docs.md for rules. Run scripts/lint-and-fix.sh to auto-fix titles over 60 chars and descriptions over 160 chars.

Version-controllable. Composable. Portable across Claude Code, Claude.ai, and (increasingly) third-party agents that speak Skills.

Custom GPT — UI-based

1. Open ChatGPT → “Explore GPTs” → “Create”
2. Paste instructions in the UI
3. Upload knowledge files (20 max)
4. Configure Actions via OpenAPI schema
5. Toggle web browsing / DALL-E / Code Interpreter
6. Publish to GPT Store

No way to version control, no file-based definition, no export/import across accounts (yet).

Gemini Gem — UI-based

1. Open Gemini → “Gem manager”
2. Name it, write instructions
3. Optionally upload knowledge
4. Save; use in chat

Extremely simple. Nothing fancy, but also limited.

Distribution and discovery

GPT Store (millions of Custom GPTs)

• Biggest catalog by far
• Revenue sharing for US Plus/Team/Enterprise creators (since 2024)
• Discovery is noisy — much low-quality content
• Featured GPTs get massive traffic

Claude Skills ecosystem

• Smaller but growing fast
• Higher signal-to-noise — curated + community
• Anthropic’s monetization launched Q4 2025 (favorable terms, 85/15 split)
• Popular Skills: coding-agent, github, research, healthcheck

Gemini Gems

• No public marketplace in April 2026
• Workspace-level sharing only
• Gemini Labs has demo Gems but no third-party distribution

For reach: Custom GPTs win. For depth and monetization: Claude Skills are catching up fast. For Gems: distribution isn’t the point — they’re productivity helpers.

Enterprise story

Area	Claude Skills	Custom GPTs	Gemini Gems
Private org-only distribution	✅	✅	✅ (Workspace)
Admin controls	✅	✅	✅
Audit logs	✅	✅	✅
Data residency	✅ (Enterprise)	✅ (Enterprise)	✅ (Workspace)
Zero-retention mode	✅	✅	⚠️ Partial

All three have maturing enterprise stories. If you’re on Claude Enterprise, Custom GPT Enterprise, or Google Workspace, each platform’s native approach works. The deciding factor is which LLM your users already use.

Who should build what

✅ Build Claude Skills if…

• You’re building developer tooling (Claude Code is the natural home)
• You need code execution, scripts, MCP integration
• You want version control + git-based workflows
• You value high-quality curation over reach
• Your users run Claude Code already

✅ Build Custom GPTs if…

• You want maximum reach (GPT Store = massive discovery)
• You want to monetize without owning infra
• Your use case is consumer or SMB-facing
• You need Actions (API integrations)
• You’re OK with ChatGPT being the only home

✅ Build Gemini Gems if…

• You need Workspace-internal helpers
• You need something fast, simple, instruction-based
• Your audience is your team, not the world
• You’re fine with limited capability ceiling

Real-world examples

Best Claude Skills (April 2026)

• Superpowers — collection of professional developer skills
• coding-agent — wrap Codex/Pi/Claude Code for sub-agent work
• github — full GitHub CLI workflows
• healthcheck — security hardening
• See our Claude Code skills guide

Best Custom GPTs (April 2026)

• Consensus — academic research GPT
• Scholar GPT — research synthesis
• Image Generator GPT — DALL-E 3 wrapper
• Code Tutor — interactive learning

Best Gemini Gems (April 2026)

• Writing Editor — in-Workspace doc edits
• Career Coach — interview prep
• Study Buddy — K-12 learning
• (All official Google Gems, no third-party marketplace)

The platform bet

Each is a bet on a different future:

• Claude Skills bet: agent workflows and developer tools win. MCP is the protocol. Skills are the distribution format.
• Custom GPTs bet: ChatGPT is the Windows of AI. Consumer reach and marketplace economics win.
• Gemini Gems bet: AI assistants belong inside existing productivity tools, not as standalone products.

If you believe in the developer + agents future → Skills. If you believe consumer reach wins → GPTs. If you believe AI is just a feature of productivity → Gems.

Verdict

For developer tools and agent workflows, build Claude Skills. They’re the most powerful format, MCP-native, version-controllable, and monetizable on favorable terms. In April 2026 they’re the highest-ceiling choice.

For consumer reach and monetization, build Custom GPTs. The GPT Store is still the only game in town for “millions of people might discover my AI product.” Revenue sharing and Actions make real businesses possible.

For internal team helpers, Gemini Gems are fine — especially if your org is on Google Workspace. Don’t build Gems for external distribution; it’s the wrong tool.

Many teams build on multiple. Core product = Claude Skills. Consumer funnel = Custom GPT. Internal team assistants = Gems. These are not mutually exclusive, and the formats are different enough that porting is non-trivial but possible.